← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 9049.e · 生成模型:GPT-6

重建优秀不等于深度唯一正确

理解等级:中级

光度损失可以借多种深度/运动组合、无纹理区域与外观歧义取得较低值。应结合独立深度参照、跨帧一致性和域测试判断。书中对监督与无监督未来优劣的判断属于当时讨论,不是由数据量多推导出的性能定理或当前方法排名。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      In this chapter we have reviewed some of the key components of a system trained to estimate depth from single images. The methods are similar to other regression problems and most of the challenge consists of identifying how to collect the data needed for training. Supervised methods are more reliable than unsupervised methods for now. Unsupervised methods rely on learning to estimate depth using the

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第43章《单图深度估计》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c043.xhtml) - 结论依据:摘录中关于重建优秀不等于深度唯一正确的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:In this chapter we have reviewed some of the key components of a system trained to estimate depth from single images. The methods are similar to other regression problems and most of the challenge consists of identifying how to collect the data needed for training. Supervised methods are more reliable than unsupervised methods for now. Unsupervised methods rely on learning to estimate depth using the